Microsoft Unveils Aurora: A Breakthrough AI for Climate and Disaster Forecasting

Amid intensifying global climate change and the growing frequency of natural disasters, the demand for technologies capable of making highly accurate forecasts is rising significantly. In response, Microsoft researchers have developed a new artificial intelligence model called ‘Aurora’, designed not only to forecast weather conditions but also to accurately and swiftly estimate various environmental factors such as air pollution and ocean wave intensities. These findings were published in a study in the Nature journal.

Aurora is classified as a “foundation model,” meaning it is trained on extensive datasets and can later be fine-tuned for specific applications. The training data comprised nearly one million hours of satellite imagery, radar information, data from meteorological centers, historical weather patterns, and prior forecasts. According to Microsoft researchers, this is the first instance of an AI weather model being trained with such a massive dataset.

Following this foundational training, Aurora was fine-tuned with smaller sets of additional data to predict specific elements like air quality, wave heights, and cyclone trajectories. When fine-tuned, Aurora outperformed existing numerical models and other AI-based methods in medium-range weather forecasting (up to 14 days), achieving better results in 91% of target areas. Notably, while traditional methods take considerable time, Aurora can generate forecasts within seconds—making it nearly 5,000 times faster. Despite the high initial training cost, researchers noted that operational costs post-deployment are very low.

Aurora accurately predicted the landfall zone of Typhoon Doksuri, which struck the Philippines in July 2023, four days in advance. Official forecasts had suggested the storm would veer away from the northern coast of Taiwan. Additionally, during the 2022–2023 season, Aurora outperformed seven major meteorological centers in tracking the paths of cyclones globally. It also predicted the devastating sandstorm that hit Iraq in June 2022 a day in advance, and at a significantly lower cost. During severe events such as Typhoon Nanmadol that struck Japan in September 2022, Aurora also estimated ocean wave heights more accurately than existing models.

To promote further innovation in meteorological science, Microsoft has made the Aurora source code and model weights publicly available. Microsoft’s MSN Weather platform is already utilizing Aurora’s AI capabilities. Researchers believe Aurora will not replace existing weather forecasting systems but will instead serve as a powerful enhancement to them.